convergence not achieved???

[Sonia Zam]

I have used modified Poisson regression models with robust error variance to estimate the prevalence ratio (PR) and to determine which demographic factors were associated with exposure to fish.

The prevalence of exposure to fish among the people who completed to survey is low, e.g. 2% of the respondents are exposed to fish. This data from a cross-sectional based study.
I have run the following command:

glm b04fish_new i.gender i.edugrp ib12.occgrp , fam(poisson) link(log) nolog vce(robust) eform

In the results it indicate that “convergence not achieved” and this only happen when I add “occgroup” variable to the model, which is a catergorical variable having 27 catogries that represent the job held by the respondents.

I am using stata14

Any suggestion on how to fix this. Thanks.

[Carlo Lazzaro]

Sonia:
have you tried running regression after regrouping -occgrp- levels in wider categories?

[Sonia Zam]

I can’t regroup because each category represents a specific occupational group that have similar job tasks. thanks

[Carlo Lazzaro]

Sonia:
far from me any intent to insist but (say) -food_prep- and -food_service- are really so different?

[Maarten Buis]

I would start with looking at tab occgrp. I suspect that many of the categories just contain too few observations. In that case you just have no choice but to do what Carlo told you (or just leave occupation out of your model).

[Joao Santos Silva]

Dear Sonia,

As Carlo and Maarten suggested, the likely cause of the problem is the combination of a dependent variable with many zeros with a set of dummies with few ones; if any of the dummies is 1 only when the dependent variable is 0, the maximum likelihood estimator does not exist. You can try to estimate the model using -ppml- (avaialble form SSC), which deals with that problem; check the help file for more information and references on this problem.

Best wishes,

Joao

[Sonia Zam]

Thank you all for your help and your comments are very useful.

[Thomas Lemma]

Hi all
I am running a poisson model using control function on STATA where the dependent variable is number of food groups against two endogenous variables.
Can I interpret the immediate results? or Do I need to go for the margins? In addition, how do I get information like F - test and R2, in ivpoisson cfunction?

Thank you

[Carlo Lazzaro]

Thomas:
welcome to this forum.
As per FAQ., please start a new thread. Thanks.

[Yasir Abdullahi]

Dear Sonia
I also had the issue. when I added the number of independent variables after poison it will work

glm b04fish_new i.gender i.edugrp ib12.occgrp , fam(poison n(3)) link(log) nolog vce(robust) reform

regards